Collecting the Evidence Flashcards
What are the basic components of a good research question ?
-Patient (i.e. define precisely whom the question is about)
-Intervention (e.g. drug treatment)
-Comparison (e.g. placebo, standard therapy)
-Outcome (e.g. side-effects, improvements on actual effects, reduced mortality)
Identify the PICO in the following scenario, and construct a research question.
• Mrs Brown is an active pensioner in general good health who has had high blood pressure for many years. The hypertension has been successfully controlled by beta- blockers.
• Her son has just been prescribed Captopril (an ACE inhibitor) to treat his hypertension.
• Mrs Brown has asked if Captopril would be better medication for her. What would your advice be?
Patient: High blood pressure, elderly
Intervention: ACE inhibitor
Comparison: Beta blockers
Outcome: Reduce BP, minimise side effects
In elderly patients, are ACE inhibitors more effective than beta blockers in controlling high BP and minimising side effects ?
Identify the main information sources when trying to answer a clinical question.
♦ Meta Analyses and Systematic Reviews (e.g. Cochrane Collaboration)
♦ Clinical Practice Guidelines (covering large disease groups and treatment strategies, e.g. SIGN, NICE)
♦ Original article containing primary research data (e.g. RCT)
Specifically, start with:
1) Cochrane Reviews
NICE and SIGN guidelines
Then use
2) MedLine (PubMed, Ovid)
Define alternative hypothesis and null hypothesis. Give an example for each.
♠ Null hypothesis- two data sets are from the same population and not different
e.g. no correlation between prostate specific antigens and prostate cancer
♠ Alternative hypothesis- two data sets are from different populations and are different
e.g. high level of prostate specific antigen is associated with cancer
What are the main categories of data ?
1) Qualitative
2) Quantitative
-Discrete (can only take certain numerical values, e.g. number of children)
-Continuous (do not have discrete steps, e.g. height and weight)
3) Categorical variables
-Ordinal (ordered categories):
•Objective: heavy/moderate/light drinker based on number of units of alcohol per wk, grade of breast cancer
•Subjective: health status questionnaire (e.g. I feel good, terrible…)
-Nominal (unordered categories): males/females, green/blue eyed…
What is the main way of testing a hypothesis, once data collection was performed (i.e. determine whether the two data sets are different from each other) ?
1) Assume the null hypothesis (i.e. two data sets are from the same population and are not different)
2) Determine the probability that the null hypothesis is correct through the P-value
As scientists, do we want P value to be high or low ? Why ?
Low, thus proving that null hypothesis is incorrect
Explain what a P-value means. What is the threshold P-value for the null hypothesis to be reasonably rejected ?
-A P-value of 0.1 means 10% chance that the null hypothesis is correct.
-If P-value < O.O5 (i.e. 5%) indicated that the null hypothesis can reasonably be rejected (i.e. there is a statistically significant difference between the two populations). If P-value is < 0.01, statistically highly significant difference. If P-value is < 0.001, statistically very highly significant difference.
How are P-values displayed on graphs ?
1) Asterisks
* for P < 0.05
** for P < 0.01
*** for P < 0.001
2) Write them out on the graph/in the legend
Define type I and type II errors (and identify the effect of each on P-value). Which of the two is more common ?
Type I- rejecting the null hypothesis when it is true (false positive)
i.e. concluding there is an effect when there isn’t (P-value is low)
Type II- not rejecting the null hypothesis when it is false (false negative)
i.e. concluding there is no effect where there is (P-value is large)
MORE COMMON
What is a potential reason for a type II error.
Because sample too small (i.e. if larger sample, would have seen smaller P-value)
Define the power of a test. What factor can influence this ?
The power of a test refers to its ability to reject the null hypothesis when it is false.
i.e. ability to detect an effect when there is one.
Sampling can affect this: need to ensure one is sampling across the population (so no bias) and sampling a large enough number of individuals to be able to detect the effect (the more individuals recruited, the smaller effect you can detect)
How value of power is deemed sufficient in clinical research ?
80 % power is deemed sufficient in clinical research
A case-control study to determine if pancreatic cancer is linked to drinking coffee. If I want 80% power to detect a 10% difference in the proportion of coffee drinkers among cases vs. controls (if coffee drinking and pancreatic cancer are linked, we would expect that a higher proportion of cases would be coffee drinkers than controls), how many cases and controls should be used sample? About half the population drinks coffee.
392 cases and 392 controls
Why is it important to overrecruit in studies ? When should power analysis be performed ?
Because some patients will drop out, which will affect sample size (and hence power).
Hence although power analysis usually done at the start, can be useful if also done at the end.
